Data-driven Estimation of Low-Power Long-Range Signal Parameters by an Unauthenticated Agent using Software Radio

Abstract

Many large-scale distributed Multi-Agent Systems (MAS) exchange information over low- power communication networks. In such scenarios, agents communicate intermittently with each other, often with limited power and over unlicensed spectrum bands that are susceptible to interference, eavesdropping, and Denial-of-Service (DoS) attacks. In this work, we consider a popular low-power, long-range communication protocol known as LoRa. Despite LoRa's high tolerance for noise and interference, it was found vulnerable to interference from particular chirp-type signals. State-of-the-art signal jamming techniques that exploit this property require the knowledge of two sensitive parameters - Bandwidth (BW) and Spreading Factor (SF). However, such information is available only to authenticated parties on the network and not to an eavesdropping adversary. We expose LoRa's vulnerability to DoS attacks by designing an intelligent jammer that surpasses the need for prior knowledge of these parameters. Exploiting a structural pattern in LoRa signals, we propose a Neural Network (NN) implementation for jointly inferring the two parameters by eavesdropping. Through simulation and experimentation, we analyze the detection vulnerability of LoRa for each combination of these parameters at various Signal to Noise Ratio (SNR) values. This work also presents a Radio Frequency (RF) dataset of LoRa signals, which is used to validate our inference model through experimentation.